TY - GEN
T1 - A Prototype-Based Few-Shot Named Entity Recognition
AU - Cao, Jian
AU - Gao, Yang
AU - Huang, Heyan
N1 - Publisher Copyright:
© 2022 ACM.
PY - 2022/3/18
Y1 - 2022/3/18
N2 - Few-shot Named Entity Recognition (NER) task focuses on identifying name entities on a small amount of supervised training data. The work based on prototype network shows strong adaptability on the few-shot NER task. We think that the core idea of these approaches is to learn how to aggregate the representation of token mappings in vector space around entity class. But, as far as we know, no such work has been investigated its effect. So, we propose the ClusLoss and the ProEuroLoss aiming to enhance the model's ability in terms of aggregating semantic information spatially, thus helping the model better distinguish entity types. Experimental results show that ProEuroLoss achieves state-of-the-art performance on the average F1 scores for both 1-shot and 5-shot NER tasks, while the ClusLoss has competitive performance on such tasks.
AB - Few-shot Named Entity Recognition (NER) task focuses on identifying name entities on a small amount of supervised training data. The work based on prototype network shows strong adaptability on the few-shot NER task. We think that the core idea of these approaches is to learn how to aggregate the representation of token mappings in vector space around entity class. But, as far as we know, no such work has been investigated its effect. So, we propose the ClusLoss and the ProEuroLoss aiming to enhance the model's ability in terms of aggregating semantic information spatially, thus helping the model better distinguish entity types. Experimental results show that ProEuroLoss achieves state-of-the-art performance on the average F1 scores for both 1-shot and 5-shot NER tasks, while the ClusLoss has competitive performance on such tasks.
KW - Few shot learning
KW - Named entity recognition
KW - Prototype network
UR - http://www.scopus.com/inward/record.url?scp=85134410099&partnerID=8YFLogxK
U2 - 10.1145/3532213.3532263
DO - 10.1145/3532213.3532263
M3 - Conference contribution
AN - SCOPUS:85134410099
T3 - ACM International Conference Proceeding Series
SP - 338
EP - 343
BT - ICCAI 2022 - Proceedings of 2022 8th International Conference on Computing and Artificial Intelligence
PB - Association for Computing Machinery
T2 - 8th International Conference on Computing and Artificial Intelligence, ICCAI 2022
Y2 - 18 March 2022 through 21 March 2022
ER -